290 JOURNAL OF LAW AND POLICY
writing style in the 15th Book of Oz is more compatible with
Thompson’s than Baum’s.”^14
There are notable differences between these studies.
Mosteller and Wallace studied a variety of possible features
before settling on a hand-picked set of thirty words (including
some rather rare words such as “direction”) chosen for their
discriminative abilities in this specific study.^15 Binongo, on the
other hand, simply used the fifty most common words in the
corpus.^16 In this volume, Stamatatos argues for the use not of
words but of character sequences;^17 we have argued elsewhere
for the use both of character sequences and word sequences.^18
Mosteller and Wallace used a form of Bayesian statistical
analysis,^19 Binongo used PCA,^20 Stamatatos uses a third
technique called “support vector machines,”^21 and we have
argued elsewhere for similarity-based nearest neighbor
methods.^22
More striking than the differences, however, are the
similarities in both the Mosteller-Wallace and Binongo studies:
the set of candidate authors was limited to only a small and
clearly defined group of people;
all candidate authors had an extensive body of unquestioned
work to compare;
(^14) Id. at 16.
(^15) MOSTELLER & WALLACE, supra note 2, at 67–68.
(^16) Binongo, supra note 9, at 11–12.
(^17) See generally Efsathios Stamatatos, On the Robustness of Authorship
Attribution Based on Character N-Gram Features, 21 J.L. & POL’Y 421
(2013).
(^18) See generally Patrick Juola & Darren Vescovi, Analyzing Stylometric
Approaches to Author Obfuscation, in ADVANCES IN DIGITAL FORENSICS VII,
at 115, 115–25 (Gilbert Peterson & Sujeet Shenoi eds., 2011).
(^19) See generally MOSTELLER & WALLACE, supra note 2.
(^20) Binongo, supra note 9, at 12–17.
(^21) Stamatatos, supra note 17, at 431.
(^22) John Noecker, Jr. & Patrick Juola, Cosine Distance Nearest-Neighbor
Classification for Authorship Attribution, PROC. DIGITAL HUMAN., 2009, at
208.